Paolo D’Ingeo - ECMWF...ECMWF Newsletter Number 77 - Autumn 1997 Paolo D’Ingeo It is with great sadness that we report the sudden death of Paolo D’Ingeo from a heart valve failure

ECMWF Newsletter Number 77 - Autumn 1997

Paolo D’Ingeo

It is with great sadness that we report the sudden death of PaoloD’Ingeo from a heart valve failure early on 5 September 1997.

Paolo D’Ingeo was born in Messina, Sicily, in November 1938.Having completed his secondary education, he entered the serviceof the ‘Home Office’ in 1959, and also began studying at MessinaUniversity in the same year. He graduated in Economics andCommerce in 1964. In 1964 he was promoted to a post in theMinistry of Defence, and two years later was again promoted andjoined the Treasury Ministry. Immediately prior to joining ECMWFhe was a Division Director in the General Inspectorate for the StateBudget, Treasury Ministry.

Paolo joined ECMWF on 1 July 1976 as Head of the FinanceSection. He held this post, which was enlarged to become the‘Finance and Supplies Section’ in 1980, until the time of his death.He also held the post of Deputy Financial Comptroller.

Paolo was a meticulously thorough and conscientious Head ofFinance. His relations with his section staff were, however, relaxedand easygoing, and he was a calm and affable colleague. Over theyears he oversaw the introduction of more modern and hi-techaccounting and banking methods, and became adept himself in theuse of the new technology.

He was, above all, dedicated to his family; his wife Gillian, and hisdaughter Magda. He was extremely proud of Magda’s educationalaccomplishments, and he had seen her launched on a highly promis-ing career; sadly he will not be able to follow this in years to come.

Our heartfelt sympathy goes out to Gillian and Magda at thistragic time.


In this issue

Editorial. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

Changes to the operational forecasting system . . . 1

METEOROLOGICAL

Seasonal forecasting at ECMWF . . . . . . . . . . . . . . 2

Summary of Technical Memorandum No. 226. . . . 9

Summary of Technical Memorandum No. 228. . . . 9

Summary of Technical Memorandum No. 231. . . 10

COMPUTING

TEN-34 and DAWN . . . . . . . . . . . . . . . . . . . . . . . 10

GENERAL

ECMWF calendar. . . . . . . . . . . . . . . . . . . . . . . . . 13

ECMWF publications. . . . . . . . . . . . . . . . . . . . . . 13

Cover

Too much rain in El Salvador damages a crop, toolittle rain in Australia parches the fields. Suchclimate variability might be predictable on aseasonal timescale - see article on page 2.

[Acknowledgements: Popperfoto (El Salvador); TheAge newspaper (Australia)]

1

Editorial

In 1994 ECMWF’s Council approved a three-yearproject to investigate the feasibility of seasonal fore-casting. The project has shown that real-time seasonalforecasts can be produced which have some skill.Experimental forecasts are now being produced threetimes a week with a coupled atmosphere oceanmodel. Some results from that project are describedin pages 2-8. Information is also available onECMWF’s web site ( h t t p : / / w w w. e c m w f . i n t ) .

Three more ECMWF Technical Memoranda aresummarised on pages 9 - 10.

There has been an explosion of interest in, and use of,the Internet in the past few years. The result hasbeen a severe performance problem as the linksbecame overloaded. To alleviate this a pan-Europeannetwork has been set up connecting national researchnetworks at megabit speeds. It is known as TEN-34(Trans European Network at 34 Mbps). ECMWF isinvolved (together with Deutscher Wetterdienst andMétéo France) in a project called DAWN to monitorthe performance and to demonstrate the value of thenew network to advanced applications such asnumerical weather prediction. A report on TEN-34plus DAWN is to be found on pages 10 - 13.

Changes to the OperationalForecasting System

Recent changes

u A number of modifications were made on 27 A u g u s t1997, mostly concerning the use of satellite data(model cycle 16r3):

1 . a revision of the use of TOVS radiances including thesimplification and improvement of bias corrections.S ATEM retrievals were replaced by the use ofradiances above 100 hPa while HIRS12 were intro-duced both over sea and land;

2 . Some quality control modifications with most impacton the way ERS-2 wind retrievals are handled.

Revisions were also made to the ocean albedo and tothe momentum transport representation in the convec-tion scheme. A correction was applied to computationsof saturation vapour pressure with effect mainly invery cold conditions.

u On 11 November 1997, a modification was made toallow the assimilation of the new RTOVS format forNOAA-14 radiances (model cycle 18r1). NOAA-11RTOVS have been assimilated since 1 December.

u On 25 November 1997, the first version of a four-dimensional variational data assimilation system( 4 D - Var) was introduced. It is based on an evolutionof the previous 3D-Var system with a 6-hour cycling.Details will be given in a forthcoming newslettera r t i c l e .

u A number of changes to the physical parametrizationscheme were introduced on 16 December 1997(model cycle 18r3):

1 . a modification in the treatment of the water vapourabsorption in the long-wave part of the radiations c h e m e ;

2 . a new method of moisture convergence closure;

3 . a new treatment in the ice fall-out in the clouds c h e m e .

Planned changes

u An improved formulation of the two-time-level semi-Langrangian scheme.

u A n e w, more accurate resolved model orography.

u Coupling of the atmospheric and ocean-wave model.

u Revised initial perturbations for the EPS, toincrease spread, particularly in the early part of theforecast range.

François Lalaurette


In the last few months there have been many reports ofanomalous weather conditions in different parts of theglobe: the droughts in Northern Australia, Papua NewGuinea and Indonesia, the associated smoke pollution inthe Malaysian and Indonesian region, hurricanes batter-ing the west coast of Mexico, abundant rain on the coastalplains of South America in places where it hardly everr a i n s , transforming deserts into blooming meadows.The mighty Yellow river, ‘China’s Sorrow’, has been particularly dry this year while Hong Kong has hadrecord rainfall ( 1 ).

The effects are not just atmospheric. While theanchovy fishing in northern Peru all but ceased, thegame fishing industry off Southern California has had abumper time, with abundant supplies of tropical fish.Enjoying the warmer than usual water along the westcoast of America, voracious mackerel (with a taste foryoung salmon) are heading north towards theWashington/British Columbian salmon fishing region.The sea level has also been changing. In parts of thewest Pacific it has dropped by as much as 40 cms whilerising by similar amounts in the equatorial east Pacific.And so the list of climate fluctuations, both atmosphericand oceanic, could go on. Although the weather alwaysseems to be anomalous somewhere, this year seems to bemore anomalous than most. Why? What is different thisyear from last?

The key lies far from our shores, in the vast expansesof the tropical Pacific, where the so-called El Niño hasreturned with a vengeance equalling and in somerespects possibly exceeding the El Niño of 1982/83,which at that time was the most violent El Niño onrecord. The damage by some estimates was as high as 10billion pounds with many thousands of people losingtheir lives in raging torrents or droughts depending onwhere they happened to live.

There is a tendency to blame any and all unusualweather conditions on El Niño. This is probably not jus-tified. Even without an El Niño there would be weatherdisasters, but there is little doubt that its effects arealmost global. Even here in Europe, half a world awayfrom the source of the climate disaster, there are possiblyEl Niño effects. Although it becomes hard to ascribe anyone disaster to El Niño, over the years, observations andmodelling studies including those at ECMWF, have gonesome way to clarify potential climate teleconnections.

One conceptual model of weather is that of a series ofevents which are (for practical purposes) unconnected:the weather next week is essentially independent of theweather this week. An example of such a model is anunbiased coin. If such a coin is tossed several hundredtimes, one would expect to find short runs of one face orother, say five heads in a row, purely by chance. If theheads are thought of as inactive weather systems, thensuch a run of heads might correspond to drought condi-tions. There is no point in seeking a physical cause for

such a run. The ‘drought’ is simply the outcome of aseries of chance processes and as such it is unpre-dictable. But weather patterns may not always be purelys t o c h a s t i c . Suppose the coin were slightly biased. Then asequence of tosses would still throw up heads and tails inpretty much random ways as before, but a more carefulanalysis might reveal that there were slightly more headsthan tails, which in our simple analogue, would corre-spond to below average rainfall. Individual weathersystems may still be chaotic, but the statistics governingthem may be perturbed in a deterministic and predictablew a y. In middle latitudes where we are living the biasfrom El Niño is likely to be fairly small, although inextreme events such as now it might be stronger. Butcloser to the Pacific, the bias will be much stronger tothe point where in some cases the coin is heads onalmost every toss. In major El Niño years, there mightbe no rain to speak of for months on end, as happened forexample in parts of New Guinea this year.

How much of this year’s climate anomalies is pre-dictable is a hot topic of research. There are reasons forbelieving that some aspects of climate variability mightbe predictable on the seasonal timescale. Firstly, thelower boundary conditions of the atmosphere such as seasurface temperature (SST) or soil moisture may have alonger memory than the atmosphere and hence be atleast partly predictable. Secondly there is a fairly com-prehensive ocean observing system now in place in theequatorial Pacific which can measure upper ocean tem-peratures from the surface to depths of 500m. And finallymodels of the atmosphere and ocean have improved. To besure these are still flawed and require extensive furtherrefinement but they are good enough to use in developinga coupled system. Since it is the slower timescale in theocean that brings predictability, any attempt to predictEl Niño, or seasonal changes in general, must involveboth atmospheric and oceanic models. So it was that theCouncil approved a programme in seasonal forecastingabout 3 years ago. The group was assembled in July ’95and seasonal prediction development began. Here wepresent results from the comprehensive coupled atmos-phere ocean model we have developed.

The atmosphere model component has a spectral reso-lution of T63 and a corresponding 1.875 degree grid forphysical processes, while vertical resolution is 31 levels.The ocean resolution in the extratropics is about 2degrees but is increased in the tropics to 0.5 degrees inthe latitudinal direction, so as to resolve the equatorialwaves which are important for El Niño. The two modelsare coupled globally and without flux corrections. Theatmospheric and land surface initial conditions aretaken from the ECMWF operational analysis.

An analysis of the ocean state is made using the forcedresponse of the ocean and assimilating surface tempera-ture analyses and all available sub-surface thermalocean data. Salinity is currently not analyzed.

Seasonal forecasting at ECMWF

2

M E T E O RO L O G I C A L


As an example of the results from our ocean analysis,figure 1a) shows the SST anomalies in March 1997 , andfigure 1b) a vertical slice along the equator of the upperocean temperature. Although panel a) shows no signifi-cant anomalies in the ocean SST, there were significantchanges occurring in the subsurface ocean as seen inpanel b). The subsurface anomalies may not alter theSST immediately or locally. Indeed these anomalies maypropagate thousands of kilometres before they alter theSST. Nonetheless it is important to carry this subsurfaceinformation in a model as it may be the propagation ofsubsurface anomalies such as these which give thesystem predictability.

The forecast strategy is to couple directly the two com-ponents of atmosphere and ocean and integrate forwardfrom the initial conditions. This inevitably leads to a driftin the climate of the coupled system, due to systematicerrors in the component models. Such biases can be cor-rected for after running a model integration by removingthe drift as estimated by a set of forecasts for an earlierperiod ( 2 ). This drift correction is included in the plots weshow later of forecast SST anomalies; it does not affectthe evolution of the coupled model itself. Extensiveexperimentation for the 1980’s and especially the ’90sshows that our system has useful predictive skill.

The forecasts we present here come from our near-real-time experimental system, where three coupledforecasts are run from each week’s ocean analysis. Theatmosphere initial data are taken from the day before,the day of and the day after the ocean analysis, in orderto create three different sets of initial conditions. Threeintegrations are used to increase our sampling of thechaotic part of the system, as will be discussed below.The forecasts are performed about two weeks after thenominal start date to allow for ocean data acquisition,and extend six months into the future.

This year is of special interest because of the develop-ment of strong El Niño conditions in the Pacific. Figure2a) shows the observed SST anomalies in December1996, when a weak cold anomaly was present in the eastPacific. Also shown is the observed anomaly in May1997, by which time the El Niño warming had appeared(Figure 2b). The forecast anomaly for May, created byaveraging the forecasts starting at different days inDecember, is shown in Figure 2c). This ensemble meanforecast is quite similar to that which was later observedto occur (Figure 2b).

To examine the temporal evolution of our forecasts, wemake use of the NINO3 index, the average value of SST inthe equatorial east Pacific in the region 5S-5N, 90-150W.

Fig 1: a) The Sea SurfaceTemperature anomaly atthe beginning of March1997. b) A vertical sectionalong the equator of thethermal anomaly in theupper ocean. Althoughpanel a shows almost noSST anomaly at this time,there is a sizeable thermalanomaly of up to 5K atabout 100m below thesurface. The contour inter -val is 1K in both plots.

a)

b)

3



Longitude

Longitude

Longitude

120°E 140°E 160°E 180° 160°W 140°W 120°W 100°W 80°W

120°E 140°E 160°E 180° 160°W 140°W 120°W 100°W 80°W

120°E 140°E 160°E 180° 160°W 140°W 120°W 100°W 80°W

60°N

40°N

20°N

0°

20°S

40°S

60°N

40°N

20°N

0°

20°S

40°S

60°N

40°N

20°N

0°

20°S

40°S

-0.5-1.0-1.5-2.0-2.5 0 0.5 1.0 1.5 2.0 2.5

Fig 2: Plot of SST anom -alies in the Pacific sector.a ) Observed anomaly inDecember 1996;b ) Observed anomaly inMay 1997;c ) Predicted SST anomalyfor May 1997 from fore -casts initiated duringDecember 1996.Contour interval is 0.5°C.The change in equatorialSST was predicted quitewell.

a)

b)

c)

4



Figure 3 shows the evolution of the Nino3 anomaly forour 6-month forecasts for start dates in a) November 96,b) February 97 c) May 97 d) August 97. There aretypically 12 to 15 forecasts on each plot. The red lineshows the observed SST. The forecasts are consistent inindicating the onset of a large El Niño in about May,although the amplitude of the event tends to be under-estimated in the longer range of the forecasts. Thespread of the forecasts starting in May 97 is larger thanfor other months and the forecasts from April 97 under-predict the development of the warming. Overall, figs 2and 3 show that our model has consistently forecast thecurrent warming very well. Although this level of skillshould not be expected on all occasions, it is encouragingthat the unusual timing of this year’s warming has beenwell captured by our system.

It is clear from Figure 3 that there are appreciable dif-ferences between the individual forecasts. One mighthave hypothesized that these differences are primarilydriven by noise in the ocean analyses, but we demon-strate here that atmospheric variability is a majorsource of the differences. It has long been realised thatthe atmosphere is highly nonlinear (chaotic) on thetimescales of weather and that individual weatherevents are unpredictable beyond a few days. It is typi-cally assumed, however, that the ocean does not reacttoo much to this high frequency variability but acts as afilter and responds only in some smoother sense.Because of the feedbacks between ocean and atmos-phere, different integrations of the coupled systemwould be expected to show some divergence of SST, butthe importance of this for eg six month forecasts hasremained unclear. Partly because of computer con-straints, the rate of divergence seems never to have beentested with a coupled forecast GCM, although estimateshave been made in other ways (3,4).

Extensive experimentation with our coupled systemhas shown that the rate of divergence, even in the trop-ical Pacific, can be surprisingly large. We illustrate thisin Figure 4, showing three forecasts initiated for 21, 22and 23 December 1996. The ocean state is the same inall three: only the atmospheric state changes. The plotsshow the predicted monthly mean temperature anom-aly in May 1997. There is a substantial spread in predicted SST between the ensemble members: all showthe overall warming, but there are differences of theorder of half to 1K over quite large areas. Differencesoutside the tropics are also large, as expected, althoughthere is some measure of both consistency and skill inthe North Pacific.

This variability in tropical SST is a result of chaoticprocesses mainly of atmospheric origin. It indicates alimit to predictive skill beyond which we cannot go sincethe differences between panels a), b) and c) arise simplyfrom different synoptic atmospheric events which areunpredictable beyond a few days. There will be otherlimitations to predictive skill: the model is not perfect,nor are the ocean initial conditions known precisely.

With better models and better observing systems, onecan anticipate that these latter limitations will bereduced in the years to come though undoubtedly noteliminated. The chaotic nature of the atmosphereimposes, however, a limit beyond which forecastscannot be improved.

While prediction of tropical SST is a sine qua non,perhaps of greater practical importance is whether wecan predict rainfall. Rainfall shows a substantially larg-er spread between individual forecasts than does SST,and this can make it difficult to establish unambiguouslywhether an ensemble mean rainfall pattern is meaning-fully different from average conditions. By taking suffi-ciently large ensembles (of order 20), statistical tests canbe used to distinguish whether an apparent signal is infact significant. For this purpose we create an estimateof the climatological rainfall distribution using forecastsfor the years 1991 - 96. Preliminary assessments of therainfall variability within this period show that signalsare often present (though rarely strong) in the northernmid-latitudes.

Figure 5 shows the ensemble-mean rainfall anomaliesfor June/July/August 1997, as predicted by our coupledsystem from initial dates in March and April. Theensemble has 27 members, and the reference climate ismade up of 114 members (19 integrations for each of 6years, all starting on the 1 April). We only plot the pointswhich are significant at the 99% level.

Several features can be seen in this figure. There ismuch more signal in the tropics than in higher lati-tudes, as expected. Nonetheless, there are several substantial features in mid-latitudes, one of the most striking of which is the wet signal over south and central Europe. There are large areas where no signalshows on this plot. Plots with a less stringent cut-off(such as 95% or 90% rather than the 99% used) showmuch more extensive areas of colour, far beyond whatmight be expected under the null hypothesis of no sig-nal. This map is therefore a conservative estimate of theregions for which a rainfall prediction might have beenmade for this year. The large and unusual El Niño ofthis summer means that this is not a typical level ofp r e d i c t a b i l i t y, however.

Even where there are apparent signals, care must beused in interpreting the map. This figure does notcontain any explicit information on the predictedreliability of the signal: the mean anomaly is shown, butnot the scatter of the individual ensemble membersabout it. As an example, a wet signal over south/centralEurope is given by more than 80% of the ensemblemembers, which is a high (but not unanimous) degree ofc o n s i s t e n c y. A dry signal over India is given by only65 - 70% of ensemble members, and so the confidence indry weather might be considered lower. The 99% signif-icance level is related to our confidence that theprobability distribution for the rainfall is shifted, not toour confidence that the actual realization of the rainfallwill have an anomaly of the predicted sign.

5



Proper verification of probabilistic forecasts willrequire large ensembles to define the forecast probabil-ity distributions, and large numbers of independentforecast periods to test statistically the relationshipbetween forecast probability and actual outcome. Thishas not yet been done. For any individual case, one cannever be sure that a match between forecast and obser-vations is not simply fortuitous. Nonetheless, the strongwet signal over Europe was unprecedented compared tothe previous six years of our test forecasts, and highlystatistically significant. Although the pattern appearssomewhat shifted compared to the observations so faravailable, we believe it likely that there is a physicalconnection with the exceptionally wet weather occur-ring across much of Europe this summer. If true, thismeans that the wet European summer weather is not inthe category of a purely stochastic event, as discussed inthe introduction, but rather is due to a predictablecause. Our results were somewhat unexpected but mayindicate that closer scrutiny of what influences Europeanclimate is warranted.

It is tempting to link the rain with the most visibleanomaly in the climate system, namely one of thelargest El Niños this century, but this would be prema-ture as there are other sources of predictability in t h e

climate system whose importance we have yet to assess.In a system as non-linear as the earth’s climate, it mayoften be a combination of factors which produce anygiven outcome. Nonetheless, the present El Niño isundoubtedly a large perturbation to the global circula-tion, and is likely to be a significant factor in many of theanomalies seen worldwide in recent months. Empiricalstudies show evidence of links between some aspects ofEuropean weather and El Niño and other factors, butcorrelations are relatively low ( 5 , 6 ). More recent studiessuggest that a clearer picture might emerge as morerefined ways of analyzing the data are pursued ( 7 ).Dynamical methods offer the best long term hope formaking seasonal forecasts because of their greatergenerality and precision: that is, because of their abil-ity to handle unprecedented situations and to treatnon-linear combinations of factors which cannot beextracted empirically from the short observationalrecords available to us. Although the current level ofmodel error reduces significantly the reliability of theforecasts, our results suggest that useful model-basedseasonal forecasting is possible. Future improvementsin models and their initialization should increase thereliability and usefulness of such forecasts in the yearsto come.

Fig 3: Plume of monthly mean SST anomalies predicted for the Nino3 region for forecasts initiated in a) November 96,b) February 97, c) May 97 and d) August 97. Three forecasts are initiated weekly and run for six months. The heavyline shows the observed values.

a)

c) d)

b)

6



120OE 140OE 160OE 180O 160 OW 140OW 120 OW 100OW 80OW

Longitude

40O S

20O S

0O

20ON

40ON

60ON

40OS

20OS

0O

20ON

40ON

60ON


Longitude

40O S

20O S

0O

20ON

40ON

60ON

40OS

20OS

0O

20ON

40ON

60ON


Longitude

40O S

20O S

0O

20ON

40ON

60ON

40OS

20OS

0O

20ON

40ON

60ON

Fig 4: Predicted SSTanomalies for May 1997,from forecasts with atmo -sphere initial conditionsfrom: a) 21st December,b) 22nd December andc) 23rd December . Contourinterval is 0.5 ºC. Despitethe identical ocean initialconditions there aresignificant differencesbetween the plots showingthe influence of differencesin atmospheric initialconditions. None-the-lessall show the same basicEl Niño development.

a)

b)

c)

7



To the man in the street, some of this might seemsomewhat remote. What he wants to know as he sits ona gloomy February night planning his fortnight’s sum-mer holiday, is whether the weather will be better inJuly than in August, in Greece than in Spain, inSnowdonia than in the Cullins. That we can not tell him.To be sure our models are still very far from perfect.With time the models will improve and the forecasts willget better, but we will still not be able to tell him. Thereality is that we live in a turbulent world. The chaoticnature of that world means that detailed forecasts for aspecific region for a short period in time are not possible.We should be able to give him a measure of the proba-bility of above or below average rain for a season, how-ever. The sophisticated holidaymaker could learn how touse these forecasts to increase the odds of a good choiceand if he took enough holidays based on our forecaststhen he should do better than random guessing.

The travel industry might be one sector of the economywhich could use seasonal forecasts. There are otherpotential beneficiaries, but at present the user communi-ty is still ill-defined. Further development within forexample the agricultural or power sector of the economyis needed in order to exploit the information in probabili-ty forecasts of seasonal evolution. One could look to thefuture where the coupled atmospheric-oceanic model is atthe core of a sophisticated system, driving in turn modelsof crop yield, aquaculture or power demand and supply.

Information on seasonal prediction, including forecasts,is now available on the ECMWF web site.

References:

1 R. Brugge, personal communication.

2 Stockdale, T. N ., Coupled ocean-atmosphere forecasts in the

presence of climate drift. Mon. Wea. Rev., 1 2 5, 809-818, (1997).

3 Kleeman, R. and A.M. Moore, A theory for the limitation

of ENSO predictability due to stochastic atmospheric tran-

sients, J. Atmos. Sci., 54, 753-767, (1997).

4 Stockdale, T., M. Latif, G. Burgers and J.-O. Wolff,

Some sensitivities of a coupled ocean-atmosphere GCM.

Tellus, 46A, 367-380, (1994).

5 Fraedrich, K., An ENSO impact on Europe? A review,

Tellus, 46A, 541-552, (1994).

6 Barnston, A.G., Linear statistical short-term climate pre-

dictive skill in the northern hemisphere,J. Climate, 7, 1513-

1564, (1994).

7 Livezey R.E., M. Masutani, A Leetmaa, H. Rui, Ming Ji,

and A Kumar Teleconnective response of the North-Pacific

American region atmosphere to large central equatorial

Pacific SST anomalies. 1787-1820 J Clim (1997).

Acknowledgements

The model integrations were carried out on a VPP300 machine

donated by Fujitsu Limited.

The Seasonal Prediction group consists of:

D. Anderson, T. Stockdale, J. Alves, M. Balmaseda and J.

Segschneider

Fig 5: Rainfall anomalies for JJA from forecasts initiated in March and April. Only statistically significant (>99%)anomalies are shown. Particularly unusual is the significant anomaly over south/central Europe.

8



A study is made of the computational stability of semi-implicit treatments of gravity-wave motion suitable foruse with two-time-level advection schemes. The analysisis for horizontally uniform reference values of tempera-ture and surface pressure, and for hybrid pressure-basedvertical coordinates. Stability requires use of referencetemperatures that are warmer than those that can beused safely with the corresponding three-time-levelscheme. The reference surface pressure should also behigher. When stable, the two-time-level scheme is damp-ing, although the largest scales are damped less than bythe three-time-level scheme if the latter uses a typicaltime-filtering. The first-order decentred averaging ofgravity-wave tendencies used in a number of semi-

There are several ways in which the sea state may playa role in the specification of the weather parameters overthe oceans. For example, ocean waves extract momen-tum from the air flow and thus the momentum transferdepends on the steepness of the ocean waves. This sea-state dependent momentum transfer has impact on theatmospheric climate (Janssen and Vi t e r b o , 1 9 9 6 ;Janssen, 1994), in particular on parameters such as thewind speed, surface pressure and wave height.

The sea state may also have impact on the ocean sur-face albedo. Essentially, the steeper the (short) oceanwaves are, the rougher the ocean surface appears for theincoming sunlight and therefore the ocean surface albe-do gets reduced. This effect was already discussed in a bynow classic paper of Cox and Munk (1954) who were ableto infer the roughness of the sea surface from pho-tographs of the sun’s glitter. In order to do so, Cox andMunk established a theoretical relationship between thesurface roughness and the ocean surface albedo, where itwas assumed that the probability distribution of thewave slope was Gaussian. It turns out that for normalincidence of the sun rays there is no impact of the seastate on the albedo while for grazing angle there is a con-siderable reduction of the ocean surface albedo by a fac-tor of two or even more. Under the latter conditions shad-owing by the crests of the waves also plays an importantrole and Saunders (1967) studied this difficult aspect ofthe reflection of sunlight by a rough ocean surface.

Summary of ECMWF Technical Memorandum 226

Stability of a two-time-level semi-implicit integration schemefor gravity-wave motion

A.J. Simmons and C. Temperton

Lagrangian models reduces the need for a relativelywarm reference temperature profile, but causes a quitesubstantial damping of otherwise well-represented low-wavenumber modes. The low-wavenumber damping canbe avoided by using an alternative, second-order averag-ing involving a third (past) time-level. For this alterna-tive averaging, an economical spatial discretization isproposed that requires no additional departure point.Phase speeds show little sensitivity to these changes informulation. All variants of the semi-implicit methodsubstantially reduce the phase speeds of the fastesthigh-wavenumber modes when use is made of the largetime-steps possible with semi-Lagrangian advection.

[Published in Monthly Weather Review 125 (1997), 600-615]


Note on the sea state dependence of the ocean surface albedo

Peter Janssen, Bernd Becker and Jean-Jacques Morcrette

In the present Technical Memorandum we discuss pos-sible consequences of a sea-state dependent ocean sur-face albedo on simulations of the climate of an atmos-pheric model. Based on the above-mentioned results ofCox and Munk and Saunders we developed for the twodimensional case (hence sunlight and ocean waves prop-agate in the same plane) an efficient algorithm thatrelates the ocean surface albedo and the roughness ofocean waves. Results from a one year simulation of theatmospheric climate by means of cy13r1 of ECMWF’satmospheric model with a sea state dependent albedoshow a small impact on the radiation budget and a mod-est impact on the net heat flux (10 Wm-2) in areas in theNorthern Hemisphere whether the net heat flux isbetween 10 and 30 Wm-2. As expected, no impact wasfound on the mean scores of a number ten day forecasts.

References.

Cox, C. and W. Munk, 1954: Measurements of the roughness

of the sea surface from photographs of the sun’s glitter. J. Opt.

Soc. Am., 44, 838-850.

Janssen, P.A.E.M. and P. Viterbo, 1996: Ocean waves and the

atmospheric climate. J. Climate, 9, 1269-1287.

Janssen, P.A.E.M., 1994: Weather and wave prediction over

the oceans. ECMWF Newsletter, 67, 3-15.

Saunders, P.M., 1967: Shadowing on the ocean and the

existence of the horizon. JGR, 72, 4643-4649.

9



In 1985 a group of four Dutch institutes, combiningexpertise in radar technology, water waves and meteo-rology, submitted a proposal to the Netherlands RemoteSensing Board for a systematic study of radar backscat-ter from the water surface. The study was aimed at thedevelopment of a physically based algorithm for the rela-tionship between radar backscatter and the surfacewind and would consist of laboratory experiments,experiments at sea from a fixed platform and theoreticalstudies of the relevant processes. The impetus for theproposal came from the expectation that in a few yearstime the ERS-1 satellite would be launched with, in itsAMI package, a C-band scatterometer intended for themeasurement of the wind speed and direction at seafrom space. The project, which was approved and sup-ported by the Board, therefore received the name Vi e r s - 1for Verification and Interpretation of ERS-1. It lastedfrom 1987 until 1994 and has now indeed resulted in the physically based algorithm for which the project was started.

The first experiments in Viers-1 were conducted in1987 in a large wind wave flume (length 100m), whereespecially the two dimensional distribution of the smallwind waves (which are mainly responsible for the radarbackscatter), and the radar directional effects of this dis-tribution were studied. An even larger wave tank (length


The Viers-1 scatterometer model

PAEM Janssen, H Wallbrink, CJ Calkoen, D van Halsema, WA Oost and P Snoeij

213m) was transformed into a wind-wave flume in 1989to study the effects of long waves on the short ones. In1990 a final experiment was performed at MeetpostNoordwijk, a research platform 9 km off the Dutch coast,to obtain field data with which to check the model.

The development of that model meanwhile had startedand a computer program was written, based on state-of-the-art knowledge of radar backscatter, wave modellingand air-sea interaction. Initially a model was made thatcalculated the radar backscatter under given conditionsof wind and sea; later the model was inverted to obtainthe wind from the radar backscatter. For operational pur-poses the model was coupled to the WAM wave model.

The crucial test for Viers-1 was a comparison of thealgorithm with the CMOD4 model, which is mainlybased on statistical fitting, using operational data. Inthis comparison it turned out that the results of Vi e r s - 1were comparable with those of CMOD4, with Viers-1 out-performing CMOD4 slightly at both low and high windspeeds, but with CMOD4 providing a somewhat closer fitin backscatter space. This may be interpreted as anappreciable success for Viers-1: the model is the first ofits kind and still has many possibilities for furtherimprovement, whereas CMOD4 is probably the ultimatethat can be obtained with a purely statistical approach.

TEN-34 (Trans European Network at 34 Mbps)TEN-34 is the first Europe-wide high speed computernetwork for the research community. The TEN-34Project consists of two elements: firstly, a pan-European34 Mbps backbone network based on IP technology. TheIP service in TEN-34 is supported by a mixture of highspeed (up to 34 Mbps) leased lines and ATM VirtualPaths which are used as leased lines. Secondly, TEN-34works with the JAMES project. This is a collaboration ofEuropean telecom operators who have established anexperimental European ATM network, to trial ATMtechnology in a pan-European context. ATM (Asyn-chronous Transfer Mode) is a connection-oriented net-work technology, based on cell switching. ATM offers thepossibility of flexible bandwidth assignment which isimportant for real-time multimedia applications.

The TEN-34 network consists of two subnets. The firstis an IP service between the Unisource home countriesSweden, Netherlands, Switzerland and Spain, withBelgium sharing the Netherlands access point via a sep-arate link to Amsterdam. Unisource also connects to the

TEN-34 and DAWN

UK and Germany. The second subnet is a data trans-mission service based on ATM, which interconnectsFrance, Germany, Italy, the UK, Luxembourg,Switzerland, Austria, Hungary and Greece. Connectionsto the Czech Republic and Slovenia are planned for later.The two subnetworks are connected at three points:London, Geneva and Frankfurt. Users get access to theTEN-34 network via their national research network.The current topology of the network is shown in figure 1,together with the line speeds (Mbps).

DAWN (Distributed Applications over Wide AreaNetworks)

Within Europe, there is a large research communityworking on the improvement of weather forecasts. Theseresearchers are located at various national weatherservices, at universities, at research institutes through-out Europe, and at the European Centre for Medium-Range Weather Forecasts (ECMWF). There is a wide-spread willingness for collaborative efforts betweenthese researchers leading to a strong requirement for

10

M E T E O ROLOGICAL / COMPUTING


intercommunication. However, many of these potentialcollaborations are data intensive and cannot be supportedby the currently available research networks.

TEN-34 is expected to provide a communication infra-structure which will allow a number of desired collabo-rative data intensive research activities to be carried outin an effective manner. There are several projects tomonitor the TEN-34 network performance for advancedapplications and to demonstrate the value of the newnetwork. One of these is DAWN.

DAWN will study the impact of the TEN-34 networkon advanced research applications within the weatherforecasting community. The three partners of the pro-ject, the national weather services of France (METEO-FRANCE) and Germany (DWD), together with ECMWF,will study, develop and test the following applications: aDistributed Model Suite, Coupled Atmosphere/OceanModels and the Modeller ’s Workbench.

The Distributed Model Suite will be used for theprediction of the spread of air pollutants. The necessarymodel framework will be developed outside of this projectby a group of researchers located at DWD, the Universityof Stuttgart, the University of Cologne and theUniversity of Karlsruhe. At this stage, DWD does nothave the computing resources required to proceed withthe development of the above model. However, DWD hasan allocation of suitable resources at ECMWF. The avail-ability of a high-speed connection between DWD andECMWF allows the ECMWF resources to be integrated

into the Distributed Model Suite, thus enabling thetimely development of the air pollution model.

Coupled Atmosphere/Ocean Models are used forseasonal forecasting activities. To obtain a better under-standing of the errors in coupled atmosphere/ocean mod-els, it is desirable to evaluate, for example, a particularatmospheric model with a variety of ocean models.However, this typically requires the porting of the oceanmodel codes. Coupling models via suitable high-speedlinks allows such scientific evaluations to be performedavoiding the often substantial migration efforts.

The M o d e l l e r’s Wo r k b e n c h is an integrated set of toolsallowing researchers to access local and remote resourcessuch as supercomputers and unique data archives,independent of their geographical location. The tools andinterfaces of the Modeller’s Workbench should present auniform, seamless environment to the meteorologist, thusallowing him to use networked resources to accomplishcomputational research without the traditional overheadof having to learn the details of remote applications andsystem configurations. A number of the required compo-nents already exist; however, since international networkconnections in the past could not support the quantities ofmodel input/output data to be moved, many interactiveuser interfaces were not designed for use over wide-areanetwork links. With the introduction of the 34 Mbps back-bone, a number of tools so far limited to local use can nowbe offered over the wide-area network.

Networking requirements & performance measurements

The selected DAWN applications have a diverse set of net-working requirements, therefore validating different aspectsof the TEN-34 network. The Distributed Model Suite movesbulk data between participating partners, requiring a highaverage bandwidth over a period of a few hours but posing nosignificant latency requirements. The Coupled A t m o s p h e r e /Ocean Models exchange bursts of data; here the latency ofthe data exchange is the critical factor. The Modeller’s Wo r k-bench is an interactive application; the network must be ableto support the chosen graphical user interfaces efficiently aswell as the bulk transfer of the model input/output data.

The network performance will be monitored using twodifferent methods:a) For each application, a set of requirements specify-

ing the average network bandwidth, acceptable vari-ance of the bandwidth, message latency, etc. will bedefined. Actual network performance will then beassessed against these parameters through the use ofsimple test programs simulating a load typical for therelevant application. Performance data will be com-piled based on samples run at periods of the dayappropriate for the anticipated use of the applications.

b ) For each application, network performance will alsobe measured in a more subjective manner: researchusers will provide feedback on the usability of theimplemented applications. This will establishwhether the observed network performance matchesu s e r s ’ requirements and expectations.

Fig 1: The current topology of the network, together withthe line speeds in Mbps.

11

C O M P U T I N G


During the project, the requirement figures used forthe tests in a) will be refined and adjusted according touser experiences from work with application prototypes.

ECMWF’s initial activities

ECMWF will implement the first part of a Modeller’sWorkbench in Phase 1 and expand it to full functionalityin Phase 2.

To provide a service for external Member State usersvia the Internet, whereby they can use resources ofECMWF without direct access to internal machines ofthe Centre, the service has to be made secure. Thus thefirst priority will be to address the security issues ofapplications run over public wide area networks. Thisrequires the design and implementation of a securitylayer as a first step. The design and implementation ofthis security layer will include:u prevention of direct routing into the Internet networku authentication checkingu identification verification

To test the security layer an ECMWF-MARS service willbe implemented as part of the Modeller’s Workbench. Inaddition, two aspects of performance can be tested by pro-viding an interactive MARS interface for external users. T h einteractive part (selection of the requested data) requiresshort response times, while the second part transfers alarge amount of data but with no critical response time.

Figure 2 shows the basic structure of the communica-tion of the remote site and ECMWF systems that will beused in Phase 1.

To guarantee widespread portability, it has beenagreed to implement the main components at ECMWFusing the programming language JAVA. There are alsosome security issues which make JAVA the preferredlanguage for the project.

First results

Monitoring the TEN-34 network to METEO-FRANCEand DWD shows the network is already capable of deliv-ering the following performance:

Round Trip Time (RTT)The first test suite measured the Round Trip Time to adestination host on each partner site. These tests pro-duced an indication of the reliability and latency of theconnection. Message latency is defined as the time torespond to a request e.g. how soon a character is dis-played after the keystroke has been made.

The results so far are (average times):ECMWF <-> METEO-FRANCE 60 msECMWF <-> DWD 30 ms

Throughput testsThe second test suite measured various aspects of net-work performance, such as:(i) the quality of the network.(ii) request/response time using a range of parameters,

e.g. the sizes of the send and receive buffers, themessage size

(iii) bulk data transfer throughput using various protocols.

Fig 2: The basic structure of the communication of the emote site and ECMWF systems that will be used in Phase 1.

12

C O M P U T I N G


Here the results are:Buffer size Average throughput between ECMWF and:

METEO-FRANCE DWD64 byte 6.0 kbits/s 12 kbits/s8 kbytes 600 kbits/s 800 kbits/s64 kbytes 980 kbits/s 5800 kbits/s

These results have to be seen in relation to the connec-tion speeds between each site and TEN-34, which are:

ECMWF – 8 Mbits/s, METEO-FRANCE – 2 Mbits/s andDWD – 34 Mbits/s.

L a t e r, the lost packet rate and the route to the destina-tion will be monitored.

Overall, the first results indicate the network is goodenough to run the advanced applications. However, fur-ther observations are needed to show if the network isstable enough to provide the required quality of serviceover a prolonged period. Heinz Richter

Feb 26 - 27 TAC Subgroup on RMDCN

Feb 23 - 6 Mar Computer User Training Course23 - 25 Feb COM1 - Introduction for new users25 - 27 Feb COM2 - Fujitsu optimisation

2 Mar COM3 - MARS3 - 6 Mar COM4 - Graphics

Mar 16 - 12 Jun Meteorological Training Course16 - 24 Mar MET1 Numerical methods,

adiabatic formulation24 Mar - 3 Apr MET2 Data assimilation & use

of satellite data20 - 24 Apr MET3 General circulation,

systematic model errors and predictability

27 Apr - 8 May MET4 Parametrization of diabatic processes

11 - 21 May, MET5 Use & interpretation of[2 - 12 Jun] * ECMWF products

May 6 - 7 Policy Advisory Committee 9th

May 11 - 12 Security Representatives meeting

* two sessions if required

May 18 - 19 Finance Committee 59th

Jun 29 - 2 Jul Workshop - Modelling and dataassimilation for land-surface processes

Jun 15 - 16 Expert meeting on EPS

Jun 17 - 18 Seasonal Forecasting Users meeting

Jun 23 - 24 Council 48th

Sep 7 - 11 Seminar - Recent developments innumerical methods for atmosphericmodelling

Sep 28 -30 Scientific Advisory Committee 27th

Oct 7 - 9 Technical Advisory Committee 26th

Oct 13 - 14 Finance Committee 60th

Nov 2 - 4 Workshop - Diagnosis of DataAssimilation Systems

Nov 9 - 13 Workshop - WGNE/GCSS/GMPP -Cloud processes in large-scale models

Nov 16 - 20 Workshop - Parallel Processors

Dec 2-3 Council 49th

ECMWF Calendar 1998

13

COMPUTING / GENERAL

Documents

Paolo D’Ingeo - ECMWF...ECMWF Newsletter Number 77 - Autumn 1997 Paolo D’Ingeo It is with great sadness that we report the sudden death of Paolo D’Ingeo from a heart valve failure